Fast background modeling object detection using PCA and temporal difference

Non-parametric density estimation method is widely used for its independence on certain data model which makes it flexible and not fixed in advance, what´s more, it is more accurate and it has no complicated mathematics prior knowledge. For this reason, non-parametric kernel density estimation background modeling algorithms are being well studied and applied by many researchers, therefore, nowadays it has become a hot area. However, its estimation is not always correct and its computation is too much, hence, inspired by the algorithm based on Bayesian theory, we propose a fast and eligible background modeling method based on PCA (principal component analysis) with temporal difference. As to us, at learning part, background modeling is done by PCA and at detection part we calculate the value between a new point to background value range produced by eigenvector and eigenvalue information. Then temporal difference is utilized to explore the relation or persistence between adjacent frames and remove noise. Last, updating background is at iteration part. Experiment results prove that it reduces calculation, deletes noise and enhances the performance at outside environment.